Post

My exposure to agentic coding has been limited but in the time I used it at a freelance client I did not see the kind of leaps in productivity others describe. The code was sometimes useful, sometimes useless, and always required enough supervision to offset the time savings.

ceej@ceej.online

I think it’s important for anyone who (completely understandably) checked out on AI the second the crypto/nft/metaverse/snake oil chuds started pushing it to realize that, in at least the domain of software development, we are very much living in a scenario where the snake oil actually works

0 reposts23 likes

I am old before my time because the pitch seems to be "you don't have to write that boilerplate code to consume one API and output to another" and my response is "but I enjoy writing that code and also paying my rent???"

My take there is that "too much boilerplate" is a useful signal that a refactor is warranted, and throwing compute at it is a very inefficient solution. Imagine if you had a legacy codebase that took too long to write code in and your solution was just to hire more engineers rather than refactor!

That said, for small bits of assistance or getting a task started, I found AI was a useful little tool. Developers do a lot of copy-pasting code from online, and a language model is well suited to doing a more flexible version of that, mapping human language inputs to code outputs.

BUT, in my own work I've refused to pay for any of these tools. The tech can be genuinely useful, but it seems completely insane to me to outsource coding itself to a cloud service with a subscription. It's bad enough we did this with CI/CD but now we're even renting the ability to write code?!

I have been assured that the local models on your $10,000 workstation are almost as good as the cloud services and will catch up soon I have given up on the objections to the flagrant abuse of source code not licensed for this kind of commercial use because the law is fake for big companies

To say nothing of the fact that they're trained on stolen code. I do wish we could get to a point of having open-weight models trained on ethically sourced data and running on local hardware, but I think a) we're a ways off from that being feasible and b) the big AI players don't want that

They're often quite well organized actually. Before LLMs, editor software had already built robust auto completion functionality using computer-readable documentation. What's harder to find is a code sample showing how to combine various functionality to match your precise use case.